Hi. I’m Professor of Politics and Affiliated Professor of Law at New York University, and Director of the Public Safety Lab. I have served as Chair of NYU’s Department of Politics and as Interim Dean of NYU’s Graduate School of Arts and Science.
My CV can be downloaded here. My email is anna [dot] harvey [at] nyu. My mailing address is Department of Politics, New York University, 19 W. 4th St., New York, NY 10012. My office number is 308.
I founded the Public Safety Lab in June 2017 to provide data science and social science support to communities and law enforcement agencies seeking to modernize and improve their criminal justice practices. For example, many relatively minor offenses are easy to detect and interdict. Many more serious offenses are much harder to detect and interdict. We may then overpunish relatively minor offenses, and underpunish more serious offenses. The Public Safety Lab works with communities and law enforcement agencies to achieve more productive allocations of public safety resources.
We are currently working with several jurisdictions on projects that include investigating: the effects of the duration of pretrial detention in over 1,000 largely rural county jails; the effects of the use of prosecutorial discretion on pre-disposition, disposition, and post-disposition outcomes; the effects of agency racial composition on public safety outcomes; the effects of appellate judge reelection and reappointment calendars on rulings in criminal appeals; the effects of assigning 911 calls involving mental health and substance abuse issues to medical response teams as opposed to law enforcement response teams; a randomized controlled trial of a platform that pushes an SMS-based survey to 911 callers, asking them to rate their experience of the police response; and the effects of urban school district calendars on the risk of sex trafficking of high school-age girls.
Through the Public Safety Lab’s Jail Data Initiative, made possible by the generous support of Arnold Ventures, we are investigating pretrial detention in approximately 1,000 largely rural county jails. Several recent papers have used the as-if random assignment of bail judges to cases to estimate the effect of pretrial detention on recidivism. These papers have reported compelling evidence that pretrial detention increases reoffending, most likely through some combination of the economic and social disruption and peer effects resulting from detention. However, these studies have all used data from the same small number of large urban jurisdictions. Very little is known about the vast majority of pretrial detainees being held in largely rural and less populated county jails. Over 1,000 of these jails post their daily jail rosters online. We are crawling these daily rosters and creating a defendant- and jail-level dataset on pretrial detention in these counties. We are also searching for counties that provide online access to criminal case and incarceration records, crawling these records, and merging them at the defendant level with the crawled jail rosters. Using these data, along with daily variation in jail capacity as an instrument, we will estimate the effects of pretrial detention on outcomes across a much broader range of jurisdictions than has previously been possible.
In the Public Safety Lab’s Prosecutorial Reform Initiative, social science and engineering teams are working with three large district attorney’s offices to a) conduct a rigorous quasi-experimental study of the causal effects of prosecutorial decision making on pre-disposition, disposition, and post-disposition outcomes; b) use the findings from the causal analysis to develop and experimentally test a decision assist tool to nudge prosecutorial decision making in the direction of reducing reoffending; and c) oversee the implementation of relatively low-cost yet effective technological solutions to support evidence-based prosecution. We expect this project to result in reliable findings about the causal effects of prosecutorial discretion on case and recidivism outcomes; an experimentally validated decision assist tool that can assist district attorney’s offices in making decisions that decrease reoffending; and the demonstration of technological solutions that can support evidence-based prosecution.
In a recent paper coauthored with Maryah Garner and Hunter Johnson, we reexamined previous findings on the effects of police department racial composition on public safety outcomes. Estimating the impact of agency racial composition on outcomes is complicated by the fact that departments can choose the racial composition of their police forces. However, many police departments in the United States have experienced affirmative action litigation designed to increase the shares of nonwhite and female police officers. We reexamined whether court-imposed affirmative action plans have impacted the rates of reported offenses and/or offenses cleared by arrest, seeking to replicate and extend work by Lott (2000) and McCrary (2007). Using a series of econometric strategies, including difference-in-differences decomposition and generalized synthetic controls, we did not find a significant effect of court-imposed affirmative action plans on the rates of reported offenses or reported offenses cleared by arrest, a finding consistent with McCrary (2007). We also considered whether unlitigated agencies change their practices due to the threat of litigation, but, like McCrary (2007), were unable to identify causal evidence of such threat effects. We suggest that future research should instead seek to identify the specific causal mechanisms linking agency racial composition and public safety outcomes.
Another Public Safety Lab project, with Sidak Yntiso, involves assessing whether appellate judges treat some criminal defendants differently as a function of their retention calendars. Recent work has indicated that trial court judges facing approaching partisan elections are more likely to engage in racial bias against black defendants (Park 2017). More generally, a large body of work has found that competitive partisan elections appear to induce more anti-defendant rulings, relative to other judicial retention institutions. This work has led many to advocate against the use of partisan elections for judicial selection and retention. Yet existing work has been constrained to cross-judge estimates, limiting the ability to draw causal inferences. In New York State, intermediate appellate judges must be elected, and re-elected, in contested partisan elections in order to be eligible for gubernatorial appointment, and reappointment, as Appellate Division judges. Using content extracted from the crawled text corpus of the approximately 38,000 slip opinions in criminal appeals heard by New York State’s intermediate appellate courts between 2003-2017, appellate judge election and appointment data sourced from the New York State Board of Elections and the New York State Judicial Screening Committee, and defendant demographic and crime data crawled and extracted from the New York State Department of Corrections’ inmate database, and probabilistically matched to the defendants in the slip opinions, we report the first within-judge estimates of the effects of both reelection and reappointment calendars on judicial votes on criminal appeals. We find that impending reappointment induces more rulings against black defendants, but not against nonblack defendants. We find no additional effect of impending reelection on appellate judge votes in criminal appeals.
In two Public Safety Lab projects on 911 calls, joint with social science and data science teams at Princeton University, we are looking at the process by which 911 calls are first triaged into medical or law enforcement responses, and then are further assigned call and priority codes by 911 call takers. These initial discretionary decisions made by call takers may have large impacts on outcomes. They may also be inflected by call taker biases that are activated by information reported by callers. Working with a large policing agency, we are sourcing several years of the original audio files of 911 callers and call takers, records of call takers’ coding decisions, Computer Assisted Dispatch records of the responses to calls, and Record Management System records of call outcomes. Using the as-if random assignment of calls to call takers, we are initially investigating three questions: 1) what is the effect of a medical response, relative to a law enforcement response, on call outcomes involving mental health and/or substance abuse issues? 2) conditional on assignment to a law enforcement response, what is the effect of call assignment to officers who have received Crisis Intervention Training (CIT) on call outcomes involving mental health/substance abuse issues? 3) is there any evidence of racial bias in the handling of 911 calls by call takers, and if so, what effects does that bias have on call outcomes?
In a related project, the social science and data science teams are conducting a field experiment of a platform that sends a one-question text survey to 911 callers whose call has been answered by a large urban policing agency. The goal of the project is to assess whether providing an opportunity for those who interact with law enforcement to provide feedback on their experience can move both civilian and officer behavior in a more constructive direction. The text survey asks callers to report a rating between 1-10 for how they feel they were treated by responding officers. If a caller responds to the survey, she is sent a follow-up link to a longer survey asking for more detailed feedback on the agency’s response. Patrol officers are given access to an online dashboard reporting not only the average survey ratings for calls to which they responded, but also average survey ratings for other officers in their unit. In the field experiment, we are manipulating several aspects of the platform, and using both survey and administrative data to estimate treatment effects. For example, we are estimating the effect of receiving the text survey on caller behavior, using administrative records of subsequent calls from that phone number. We are also estimating the effect on officer behavior of receiving access to the survey ratings, using both pre- and post-treatment survey ratings and administrative records of the calls to which an officer responds.
In another project from the Public Safety Lab, we are using data science tools to try to identify the presence of sex trafficking in online ads for and reviews of sex providers. We are first extracting structured data from the texts of the very large corpus of online sex ads and reviews of sex providers posted between 2010 and 2017, merging that structured data with information on sex trafficking investigations sourced from law enforcement agencies, and looking for features of ads that are predictive of the likely presence of sex trafficking. We are also looking for telephone numbers in online sex ads that are systematically more likely to be posted on days when high school-aged girls are out of school, but adults are typically at work (e.g., teacher “planning days”). Our working hypothesis is that these ads are likely to be posted by or on behalf of minor girls, potentially by traffickers. Using these ads as “ground truth,” we are then looking to identify features of these ads that can be used to predict the presence of potentially trafficked minor girls in the full corpus of ads and reviews.
In my non-Public Safety Lab work I am currently co-authoring a casebook on judicial decisionmaking (appropriately entitled, Judicial Decisionmaking (West Publishing, 2019)) with Andrew Martin (Washington University St. Louis), Tom Clark (Emory), Maggie Lemos (Duke Law), Allison Larsen (William and Mary Law), and Barry Friedman (NYU Law). The casebook integrates both social science and legal approaches to understanding how judges decide cases.
Another current project, joint with Emily A. West (Political Science, University of Pittsburgh), investigates discrimination in public accommodations. Identification difficulties have to date precluded the estimation of causal effects from statutes prohibiting discrimination in public accommodations. We leverage the U.S. Supreme Court’s 1883 strike of the public accommodations provisions in the Civil Rights Act of 1875, along with ex ante variation in state-level statutes, to identify the impact of a federal statute protecting access to public accommodations. Using repeated geo-located medical exams of Union Army and U.S. Colored Troops veterans, and a series of geographic regression discontinuity and placebo designs, we find that the Court’s ruling led to large relative weight losses for USCT veterans in states without state-level public accommodation statutes. These findings suggest that, despite popular skepticism about the importance of discrimination in public accommodations, this form of discrimination in fact has material negative impacts on the well-being of its victims, and that statutes prohibiting such discrimination can mitigate these impacts.
A third project investigates the causal impact of money in elections. One article in this project uses the Supreme Court’s ruling in Buckley v. Valeo (1976) to identify the causal impact of removing state limits on campaign spending. The Supreme Court’s campaign finance jurisprudence rests on a distinction between spending restrictions (generally struck) and contribution restrictions (often upheld). In Buckley v. Valeo (1976), the case originating this distinction, the majority rejected an “anti-distortion” rationale for spending restrictions, claiming that campaign spending is merely an effect of candidate support, not a cause of candidate support. If this claim is true, then removing restrictions on campaign spending should have no discernible causal impacts. This article tests the Buckley majority’s empirical claim using its own ruling, which struck limits on campaign spending in state elections in 26 states. Estimates consistently suggest that the Buckley-induced removal of state limits on campaign spending led to increased Republican voteshares, increased Republican candidate entry, and decreased Democratic candidate entry in state legislative and gubernatorial elections in states affected by the ruling, and to both increased Republican House voteshares and the election of more conservative freshman Republican House incumbents in states both affected by the ruling and holding concurrent federal and state elections. These findings suggest that the rationale for the core distinction in the Supreme Court’s campaign finance jurisprudence has little empirical foundation.
Another campaign finance paper, joint with Taylor Mattia (PhD student, NYU Department of Politics), looks at the causal impact of Citizens United v. FEC (2010) on legislators’ preferences. Recent work has suggested that the Supreme Court’s ruling in Citizens United (2010), eliminating restrictions on independent spending in elections, increased the probability of election of Republican state legislative candidates (Klumpp et al 2016). Left unexplored has been whether the Court’s ruling in Citizens United not only increased the number of Republican state legislators, but also induced the movement of state legislators’ preferences in a more conservative direction, net of any effects on Republican candidates’ probabilities of election. We attempt to distinguish these electoral and preference effects of Citizens United. Estimates consistently suggest that the Citizens United-induced removal of state restrictions on independent spending led not only to increased probabilities of election for Republican state legislative candidates, but also to larger within-district increases in the conservatism of state legislators’ preferences in formerly Democratic districts electing Republican state legislators post-ruling. These estimates, which are robust to a series of matching and placebo exercises, may provide support for the claim that an increased presence of money in elections has contributed to the increased conservatism of Republican elected officials.
Several of these projects involve applying the tools of causal inference to historical data. In a paper prepared for a special issue of Public Choice on Causal Inference and American Political Development, I explore the application of regression discontinuity (RD) designs to three questions of interest to researchers in the subfield of American political development (APD). APD scholars have long been interested in questions related to the development of “state capacity” in the United States, or the growth of a salaried and merit-based federal bureaucracy capable of competently administering programs of social provision. I illustrate how RD designs can be used to investigate the impacts of the relative absence of federal state capacity during the 19th century; of the subsequent growth of a professional salaried civil service around the turn of the 20th century and beyond; and of the resulting growth of the presence of the administrative state in Americans’ daily lives. I first illustrate the use of a geographic RD design to estimate the causal impacts of a Reconstruction-era federal civil rights statute during the period prior to the development of significant federal state capacity. Second, I explore the possible causes of the late 19th century decline in the use of monetary rewards to motivate civil servants through the use of a population-based RD design to estimate the causal impacts of financial incentives on law enforcement effort and civilian compliance. Third, I illustrate an opportunity to test claims about the impacts of the growth of the “carceral state” through the use of a modified resource constraint RD design to estimate the causal impacts of police deployments on a variety of outcomes.
In The Civil Rights Cases (1883), Buckley v. Valeo (1976), and Citizens United (2010), unelected judges struck federal statutes enacted by legislative majorities; legislative supermajorities are required to overturn these rulings. In another project I investigate the historical origins of this form of “entrenched” judicial review. Among those current democracies that were former colonies, the presence of entrenched judicial review is strongly associated with pre-colonial histories of marked inequality, a finding consistent with the hypothesis that entrenched judicial review was adopted at least in part to preserve pre-colonial inequality from legislative redistribution.
Yet unelected judges are not necessarily unresponsive to legislative preferences. In the U.S., federal judges serve only on the condition of good behavior, and congressional majorities control judicial salaries, budgets, and jurisdiction. In A Mere Machine: The Supreme Court, Congress, and American Democracy (Yale University Press, 2013), I reported evidence indicating that, even in constitutional cases, the U.S. Supreme Court defers to congressional preferences, in particular to the preferences of majorities in the House of Representatives (the chamber that originates both impeachment and appropriations actions). To view an interview about A Mere Machine on CSPAN’s Book TV, click here.
In order to generate the findings reported in A Mere Machine, several methodological challenges had to be addressed. These challenges included constructing an objectively defined measure of the ideological direction of Supreme Court judgments, the initial work for which was done jointly with Michael J. Woodruff (former NYU Politics PhD student). They also included addressing the selection bias in the Court’s docket, the early work for which was done jointly with Barry Friedman (NYU Professor of Law) (here and here). Other work on the responsiveness of the Supreme Court to congressional preferences may be found here and here.
In earlier work I investigated the proposition that partisanship may be modeled as a social convention, providing social benefits to in-group members (and social penalties to out-group members). Electoral laws that make partisan acts more public (e.g. party registration) or less public (e.g. secret ballots; party primaries rather than caucuses) will then affect the ability of neighbors to coordinate on local party social conventions, a hypothesis for which there is empirical support.
This work on partisanship as a social convention grew out of my first major project, which investigated the competition to mobilize women’s votes after constitutional female suffrage. In Votes without Leverage: Women in American Electoral Politics, 1920-1970 (Cambridge University Press, Series on the Political Economy of Institutions and Decisions, 1998), I suggested that, if there is a significant social component to turnout and partisanship, then competition to mobilize votes through social networks will be much like competition to mobilize consumers of services offered through networks (e.g., telephone networks). Markets for networked services are generally marked by imperfect competition, with early entrants having significant advantages over later entrants. Likewise, the decision by the National League of Women Voters (the former female suffrage organization) in 1923 to cede the market in women’s votes to the party organizations may have followed from its necessarily late entry into this imperfectly competitive market. Votes Without Leverage built on articles published here and here. A post on Vox in honor of the 100th anniversary of Jeannette Rankin’s 1917 swearing-in as the first woman to join the House of Representatives may be found here.